math, cmd/internal/gc, cmd/7g: enable SQRT inlining, add assembly implementation

benchmark                 old ns/op     new ns/op     delta
BenchmarkSqrt             474           16.5          -96.52%
BenchmarkSqrtIndirect     476           38.1          -92.00%
BenchmarkSqrtGo           484           483           -0.21%

Change-Id: I5ad0132feda0d3275a884523b9e79d83db4fc726
Signed-off-by: Shenghou Ma <minux@golang.org>
Reviewed-on: https://go-review.googlesource.com/8465
Reviewed-by: David Crawshaw <crawshaw@golang.org>

5 files changed, 29 insertions, 16 deletions

diff --git a/src/cmd/7g/gsubr.go b/src/cmd/7g/gsubr.go
index 2d2bdb7ca4..cea1e9205f 100644
--- a/src/cmd/7g/gsubr.go
+++ b/src/cmd/7g/gsubr.go
@@ -938,6 +938,9 @@ func optoas(op int, t *gc.Type) int {
 
 	case gc.ODIV<<16 | gc.TFLOAT64:
 		a = arm64.AFDIVD
+
+	case gc.OSQRT<<16 | gc.TFLOAT64:
+		a = arm64.AFSQRTD
 	}
 
 	return a
diff --git a/src/cmd/7g/prog.go b/src/cmd/7g/prog.go
index 2763e27a1e..733c405fac 100644
--- a/src/cmd/7g/prog.go
+++ b/src/cmd/7g/prog.go
@@ -60,18 +60,19 @@ var progtable = [arm64.ALAST]obj.ProgInfo{
 	arm64.ACMP:   {gc.SizeQ | gc.LeftRead | gc.RegRead, 0, 0, 0},
 
 	// Floating point.
-	arm64.AFADDD: {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFADDS: {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFSUBD: {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFSUBS: {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFNEGD: {gc.SizeD | gc.LeftRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFNEGS: {gc.SizeF | gc.LeftRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFMULD: {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFMULS: {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFDIVD: {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFDIVS: {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
-	arm64.AFCMPD: {gc.SizeD | gc.LeftRead | gc.RegRead, 0, 0, 0},
-	arm64.AFCMPS: {gc.SizeF | gc.LeftRead | gc.RegRead, 0, 0, 0},
+	arm64.AFADDD:  {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFADDS:  {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFSUBD:  {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFSUBS:  {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFNEGD:  {gc.SizeD | gc.LeftRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFNEGS:  {gc.SizeF | gc.LeftRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFSQRTD: {gc.SizeD | gc.LeftRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFMULD:  {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFMULS:  {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFDIVD:  {gc.SizeD | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFDIVS:  {gc.SizeF | gc.LeftRead | gc.RegRead | gc.RightWrite, 0, 0, 0},
+	arm64.AFCMPD:  {gc.SizeD | gc.LeftRead | gc.RegRead, 0, 0, 0},
+	arm64.AFCMPS:  {gc.SizeF | gc.LeftRead | gc.RegRead, 0, 0, 0},
 
 	// float -> integer
 	arm64.AFCVTZSD:  {gc.SizeD | gc.LeftRead | gc.RightWrite | gc.Conv, 0, 0, 0},
diff --git a/src/cmd/internal/gc/walk.go b/src/cmd/internal/gc/walk.go
index a0a29d35ac..c845f783e5 100644
--- a/src/cmd/internal/gc/walk.go
+++ b/src/cmd/internal/gc/walk.go
@@ -624,7 +624,7 @@ func walkexpr(np **Node, init **NodeList) {
 
 		if n.Left.Op == ONAME && n.Left.Sym.Name == "Sqrt" && n.Left.Sym.Pkg.Path == "math" {
 			switch Thearch.Thechar {
-			case '5', '6':
+			case '5', '6', '7':
 				n.Op = OSQRT
 				n.Left = n.List.N
 				n.List = nil
diff --git a/src/math/sqrt_arm64.s b/src/math/sqrt_arm64.s
new file mode 100644
index 0000000000..9861446eb9
--- /dev/null
+++ b/src/math/sqrt_arm64.s
@@ -0,0 +1,12 @@
+// Copyright 2015 The Go Authors.  All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include "textflag.h"
+
+// func Sqrt(x float64) float64
+TEXT ·Sqrt(SB),NOSPLIT,$0
+	FMOVD	x+0(FP), F0
+	FSQRTD	F0, F0
+	FMOVD	F0, ret+8(FP)
+	RET
diff --git a/src/math/stubs_arm64.s b/src/math/stubs_arm64.s
index 2ffd2289b8..eea81e9241 100644
--- a/src/math/stubs_arm64.s
+++ b/src/math/stubs_arm64.s
@@ -84,8 +84,5 @@ TEXT ·Sin(SB),NOSPLIT,$0
 TEXT ·Cos(SB),NOSPLIT,$0
 	B ·cos(SB)
 
-TEXT ·Sqrt(SB),NOSPLIT,$0
-	B ·sqrt(SB)
-
 TEXT ·Tan(SB),NOSPLIT,$0
 	B ·tan(SB)